[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Time to optimize :operation-handled-p and :send-if-handles 
I investigated :OPERATION-HANDLED-P (OHP) and :SEND-IF-HANDLES (SIH) a
bit further and discovered that a LISPM spends a fair portion of its
life inside them, and by optimizing these methods you can make dynamic
windows and the LISPM overall seem to run a bit faster. Here's what i
1. First i monitored OHP and produced a histogram of how often OHP was
called on a particular operation. Here is a typical example collected
over at least a half hour:
The histograms typically have about 50 operations in them.
2. I then did the same thing for SIH. Here is an example for only 60
seconds on a machine loading a system:
Although SIH is faster than OHP it does a WITHOUT-INTERRUPTS.
3. Obviously, SIH is being called a huge number of times per minute,
presumably by the window system.
4. I poked around and noticed that DW:DYNAMIC-MOUSE-WINDOW did both
an SIH and and OHP of some of the methods in the above histograms. So
i timed it. On my 3640 a single call takes 7.85e-3 sec, and it is
called between around 10 times a second (median of 3.37 4.44 9.88
51.98 57.5 calls/sec) so something on the order of 7% of every second
is spend inside DY:DYNAMIC-MOUSE-WINDOW.
5. Luckily, by writing a couple of methods you can make
DYNAMIC-MOUSE-WINDOW 60 times faster.
6. I think we should take a serious look at such performance issues.
Who knows were we'll find some improvements?